Research On Collaborative Parameter Identification Method Of Engine System For Virtual Calibration

Model-based virtual calibration technology plays an important role in accelerating the engine development cycle and reducing the development cost in this age of intelligence.The core of virtual calibration is to use the limited experimental resources to fast establish a high-precision engine model to replace part of the experiment.Modeling methods based on data calibration depend too much on the number of experimental points.The modeling method based on the physical process hardly depends on the amount of data and can reflect multiple working process parameters and output results simultaneously.However,problems such as the overcomplicated architecture,difficulties in parameter identification,and difficulties in considering the output accuracy within the range of all working conditions still exits in the process of mechanism modeling.To solve the above problems,a general engine system model parameter identification method is developed in this paper,and the required experimental set is optimized according to the data characteristics of the model to reduce the cost and the development period.The engine system model was built based on GT-Power platform,and a call architecture between MATLAB and GT-Power platform was established.The automatic identification of model parameters is realized by the intelligent parameter identification algorithm program written in MATLAB platform.This thesis took a 1.3L engine as an example,the parameter identification process of the engine system model is divided into two steps:parameter identification of sub-models and parameter co-optimization of the whole engine model.The results showed that the method based on parameter fitting or intelligent optimization can complete the parameter identification of each sub-model.A total number of 43 parameters were identified,and the~2 values of other sub-models were all above 0.9 except the friction model.On this basis,the key parameters in the system model were co-optimized by the Bayesian Optimization algorithm,and the percentage of operating points with the error of key output results such as air intake,torque and BSFC less than 10%was increased from 46.6%to 97.9%.In order to verify the effectiveness of this method,different numbers of points were selected from all 2660 points for parameter identification based on the uniform point selection method,and the accuracy of the identified model was verified through all2660 experimental points.The results showed that only when 100 experimental points are used for model parameter identification,the goal that more than 95%of the operation points could reach an accuracy higher than 90%can be reached.In order to further reduce the number of required experimental points and improve the accuracy,model sensitivity analysis and feature extraction were used to assign weights to experimental points to select the most important experimental points for the output results of the system model and carry out targeted experiments.It is found that the high-weight experimental sites are mainly concentrated in the low-load area and the medium-high load area near 3000r/min.The 50 experimental points with the highest weight value were selected for parameter identification and the model accuracy of more than 95.9%operating points could be higher than 90%.The generalization ability of the system model identification method and weight assignment method is verified by the model parameter identification process of another 1.4L engine,which proves the universality of the method to a certain extent.